I am a high school teacher who has taught coding for > 30 years, robotics ~7 years and machine learning using TensorflowJS and now EdgeImpulse for a few years.
Projects I am working on:
My latest robotics course, with some ML uses the Arduino PortentaH7 with LoRa Vision Shield, 320x320 Grayscale OLED and the MRK IMU ( for Acceleration data). My teaching github with videos is at https://github.com/hpssjellis/maker100
.
.
I received an $8000 grant to get microcontroller tech equipment into the hands of Indigenous students enrolled at my school.
.
.
I have made an Arduino MBED examples library called “Portenta Pro Community Solutions” just search for “community” in the Arduino IDE library manager. You should find lots of example robotics code to get anyone started with microcontrollers.
.
.
I am designing a pcb to breakout all of the 160 pins of the Portenta high density connector for students to easily use, since the headers are arraigned exactly like the HD connectors.
.
.
Presently I am working on trying to simplify the coding for after students load a machine learning problem.
This image is a FOMO example (faster objects more objects) 72ms analysis of a line, My problem is now how to get the toy car to understand from the data that it needs to turn to the right. For me this issue looks easy but is actually a pain to solve.
I kind of want to take the output from that model and put it into a different model to make the driving decisions, which the dual core Portenta could probably do, but it would slow everything down. I think, for racing, each car would have it’s own line and emergency devices (range finder or radio override), but we are not there yet.
An extra issue is the boxes get taller as the lines get straighter, see image. This should make driving decisions to go straight and faster easier, but I am not yet there with the code. Also note this image shows my Portenta PCB which is partially working (The long traces are messing with the USB-serial).
Perhaps you could take a look at the AWS Robomaker website to see if there are any clues that can help your robot navigate the line with a better model and algorithm to make it smooth and seamless.
